DeViSE: A Deep Visual-Semantic Embedding Model

   Abstract